Machine Learning Performance 13
- TVM FFI Deep Dive: Shipping Native Kernels Without Tying Them to One PyTorch Build
- NVSHMEM Deep Dive: One-Sided GPU Communication, and How It Differs from NCCL
- Where gpu_memory_utilization Actually Goes: vLLM's Memory Budget, Line by Line
- Fused Linear Cross-Entropy: The Math, the Memory Wall, and Two Ways Around It
- PyTorch DDP Deep Dive — Initialization, Buffer Broadcast, the Reducer, and the torch.compile Story
- How vLLM's `torch.compile` Backend Handles Dynamic Batch Sizes — A Code Walkthrough
- Torch.compile Deep Dive II — Inductor Codegen and Buffer Lifetimes, End to End
- SGLang Deep Dive: How Engine.generate() Boots and Runs
- vLLM v1 Deep Dive: How LLM(...) and the Server Boot and Generate
- CUTLASS Deep Dive: From CuTe Layouts to Blackwell tcgen05
- NCCL Deep Dive: From P2P/CUMEM to torchrun Init
- From FlashAttention to FlashAttention 2
- Torch.compile 101 - From Python Function to Triton Kernel